Attack Simulation
June 29, 2026 · View on GitHub
This document describes the security model for the SIEM forwarding feature in AdversaryGraph v5.
What SIEM forwarding does
SIEM forwarding sends synthetic telemetry events from the AdversaryGraph backend to an HTTP(S) or syslog collector. The events are AI-generated or predefined kill-chain telemetry records formatted as JSON, CEF, or plain syslog. They are NOT live attack traffic.
Real vs. synthetic telemetry
| Property | Real lab telemetry | Synthetic (AI-generated) telemetry |
|---|---|---|
| Source | Actual lab endpoint or web target | AdversaryGraph backend (AI-generated or templated) |
| Authenticity | Produced by the OS, process, or network stack | Generated by LLM or predefined template |
| Use case | Verify actual attack produces expected events | Test SIEM rule logic without a live lab |
| Required lab infra | Yes — configured lab target | No — runs directly from the workbench |
| Analyst validation required | Yes — confirm rule fires on real traffic | Yes — confirm rule logic matches real behavior |
| Safety | Runs in isolated lab environment | No execution; purely data forwarding |
Analysts must not treat synthetic telemetry as proof that a detection works against real attacks. Synthetic events validate rule syntax and field mapping only. Full validation requires correlation with a real lab run.
What is NOT stored
- SIEM destination URLs are saved to the recent-destinations list for UI convenience but are not treated as secrets
- Bearer tokens, passwords, and API keys entered in forwarding authentication fields are NOT stored server-side
- Synthetic event content is NOT sent to any third-party service — only to the destination the analyst configures
Authorized targets only
The SIEM forwarding feature is intended for use against analyst-controlled SIEM collectors. Forwarding synthetic telemetry to unauthorized destinations is misuse of the tool.
Confidential data warning
Do not include confidential, customer-sensitive, classified, or internal IOCs or report content in AI-generated attack scenarios. The scenarios are passed to the configured LLM provider. Use a private or local LLM gateway if input data is sensitive.
Deployment note
The default Compose profile does not expose SIEM forwarding to the internet. If the AdversaryGraph API is deployed behind an internet-facing proxy, ensure the SIEM forwarding endpoint is access-controlled at the proxy layer.